WriteReader AI Assessment

WriteReader enables K5 students to become published authors by using multimodal book creation features through the use of text, audio & images. The tool motivates the student to write interest-based books while increasing their literacy skills. WriteReader is a web-based platform accessed through any device/browser. Students can practice all language domains while creating books. Optimized to use with ELL and struggling learners.

Our AI Assessment Module will provide teachers with automated feedback on students' writing development and will be an add-on to the existing WriteReader tool. With over 5M books created, we have gathered unique data about student's early writing skills and development.

Here is a short video on how books are created with WriteReader How to Create Books

WriteReader empowers teachers with an easy-to-use tool for implementing evidence-based reading and writing instruction. With minimal planning and quick steps, WriteReader can supplement any curriculum across all content areas while motivating students to become confident, independent writers and readers. Here is an overview of our Science of Reading book templates Structured Learning Approach with Science of Reading

Teachers can easily set up their classrooms and provide feedback to the students. Teachers can adjust the literacy features down to the student level which enables the teacher efficiently to differentiate instructions and maximize the likelihood of students becoming successful authors given their current skill level.

AI-enabled assessment 

We are now implementing AI and large language models to provide students with support and feedback on their writing. At the same time, we will provide teachers with an overview of how students are progressing with their writing skills and which students need the most help. We are currently in the pilot phase and will be testing our solution with +10 schools in the US. 

Unique data to support AI assessment

We collect data when students write and create their own interest-based and multimodal digital books. We collect the following types of unique data.

Emergent writing (text and invented spelling created by students who are about to learn how to read and write or struggling in reading and writing)

Quantity - the number of words written by the student

Complexity - reading ease based on grade level Fry readability score 

Variation - Writing vocabulary that the student uses when writing books

Precision - Spelling accuracy 

The above parameters are used in our AI assessment and will be presented to the teachers so they get an overview of individual student's development and on a class level. All above literacy parameters have been carefully selected in cooperation with literacy coaches, reading intervention specialists, and literacy professors from Marshall University, WV

Individual student AI assessment overview mockup

Class-level AI assessment overview mockup

This data provides unique insights into student writing and will provide teachers with an overview of students who need the most help and what kind of intervention is needed. 

WriteReader enables students to practice all language domains (reading, writing, listening, speaking) which have made our tool a “go-to tool” for teachers, schools, and districts allowing them to implement our tool for fast, easy, and impactful literacy acquisition.

Our tool has been selected by the largest school district in Maryland (Montgomery County Public Schools) to support its FARMS schools to close the achievement gap that increased during COVID. WriteReader has been successful in closing the literacy gap as recognized by MCPS and we have continued our relationship for the past 4 years. Students with special needs and students from lower socio-economic backgrounds have particularly gained benefits from becoming proud authors. Creating books is quick, easy, and engaging.  Our multimedia book creation tool supports assistive technology like text-to-speech, speech-to-text, speech synthesis, and phonics support in order to accelerate the literacy development of the students.

Formative AI assessment 

The AI assessment will allow students to be automatically assessed based on their writing of interest-based. Teachers will know what kind of future intervention is needed.

Our data is unique as we can see patterns in how students write in different stages ranging from emergent, transitional, and fluent writing. We have now developed our AI Assessment module to equip the teachers with data on how the students are progressing and highlight the parts and phases where students need the most help. 

Time and cost-saving

Our AI assessment module will allow teachers to support their students even more as it will save teachers time from traditional assessment to automated assessment while the students are writing meaningful and interest-based books based on their existing curriculum. Our goal is to create a solution where some traditional assessments can be replaced by our automated AI assessment which will save teachers time allowing them to focus more on the students. 

AI assessment will help students with learning disabilities

Students with writing and reading difficulties need tasks that are meaningful and also provide them with opportunities to feel successful. The writing assignment must be well-defined, scaffolded, and supported by a tool with features that can meet the student’s needs during each writing phase. Many educators find WriteReader the right tool for their students. Our AI assessment will help teachers save time from traditional assessment and will give teachers time to focus on the students while they get key insight into where they need the most help. 

AI assessment to support ELL students 

We have seen an uptake from ELL teachers where districts in NC have purchased WriteReader district licenses for their ELL teachers. (Rowan Salisbury District) ELL teachers are using WriteReader with their ELL students because it enables them to tell their stories while practicing all four language domains. We will test our AI assessment module with ELL teachers and they will provide us with feedback on how our AI assessment module can support their work.

Our co founder Babar Baig has a minority background and has experienced the difficulties that can affect learning outcomes. We have always had access to underserved communities. We have conducted multiple research projects with the focus on learning outcomes for struggling readers and writers. Today, we work with US districts that represent FARM schools and our focus is to provide quality education to all students. 

Team lead, Babar Baig has conducted several sessions with refugee after-school programs enabling students to tell their refugee stories while learning to read and write. 

Our development team includes educators, software engineers, and AI experts from diverse backgrounds. This inclusion fosters a broader perspective during the tool's development and helps in understanding the unique challenges faced by marginalised student groups.

We actively engage with community stakeholders, including teachers and education experts from minority and low-income backgrounds amongst others, to gather insights and feedback on the tool's functionality and its impact on different student groups.

We founded WriteReader intending to create a motivating creation tool with documented learning value. Our founder, Janus Madsen has +16 years of experience in teaching in elementary schools and has been awarded the best teaching resource by the MOE in Denmark. Additionally, the team that will work on the upcoming scoring module will be co-founder Babar Baig who will lead the Business development efforts in the US, Lasse Sørensen, who is our data scientist employee and a master's student from the Technical University of Denmark, specializing in AI and data science.

We have participated in +10 research projects over a period of 7 years and worked with 7 different research organizations including WestEd in the US. In 2018, WriteReader was awarded a grant by the Danish Innovation Fund to automatically score students' literacy skills with AI using our unique data. We partnered with The Danish Technical University to develop the LLM (DTU) and the Danish pedagogical university to track students' literacy skills. This project is considered unique and one of the largest conducted on a global scale.

For the AI Assessment scoring module related to this proposal, we are building on our partnership with DTU and using their expertise combined with Marshall University (WV) literacy professors.

WriteReader was selected as the only non-US company to participate in the first Edtech cohort of Intel Education in 2015 and awarded an Early learning grant from NewSchools venture in 2018 recognizing our work with elementary school students.

500k students have actively used WriteReader in the last 12 months with an average user session of +31 minutes. We target teachers, schools, and districts directly through our SaaS model. Students have used our book-creating tool to create +5M books. We grew 25% last year and expect to grow 30% in 2024.

We have users in all the states in the US and 50% of all elementary schools in Denmark use our tool for language acquisition. Teachers use our tool across 40 countries. We work with educational publishers around the world including the largest in the Nordics.

We work with the largest school districts in MD, MT and ND. We have teachers who use WriteReader on a daily basis in all the states in the US. In the past 12 months, we have had 150,000 users in the US with an average user session of +31 minutes. We are about to pilot our AI Assessment Module in 20 schools across 4 states. We have a local presence in the state of Illinois (WriteReader Inc.)

Our main markets are North America and The Nordics. We currently support 14 languages, and our tool is used in classrooms in more than 40 countries. Once we implement our AI scoring module, we will directly and instantly reach thousands of teachers where they can start to see the student data, get the overview of progression, and start the needed intervention.

We have concluded 10+ studies with researchers to prove learning outcomes in the past 7 years. We are collecting data and using our own AI models when the students write interest-based digital books. This unique data allows us to find patterns in how students develop their writing. The aim is to equip the teachers with data on how the students are progressing and highlight the parts and phases where students need the most help.

We are developing the AI-scoring module for early writing along with related writing tasks. We will Integrate the module/features into our existing WriteReader platform. We plan to test the AI module in US schools and plan to fully test and scale in the next 12 months. Pilot agreements have been made with schools in the US. Our AI Assessment Module will provide teachers with feedback on students' writing development and will be an add-on to the existing WriteReader tool.

With over 5M books created, we have gathered unique data about student's early writing skills and development. Through one of the world's largest studies of emergent writing, we followed 2,200 primary school students in over a 3 1/2-year period. We used this unique data and insights to develop our own data-driven AI text proposal model and scoring module in cooperation with leading researchers from the Technical University of Denmark and the Danish School of Education.

Based on our successful results above we are currently doing a pilot project for the English language with researchers from Denmark and USA. The main goal for the pilot is to find 4-5 quantitative parameters in early writing. There are measurable indicators for students' writing development - such as quantity/word count, complexity/text readability, variation/vocabulary, and precision/spelling. This work is based on existing knowledge in the field and feedback from researchers, reading specialists, and teachers.

The goal is to automatically assess students' writing skills and provide real-time feedback to the students in the English language (WriteReader AI Assessment Module). We frame the problem of automatic scoring of students’ writing skills as a machine translation problem, where we translate the student’s early-stage writing into conventional writing with the BART sequence-to-sequence model. The student's writing skills and development are then assessed/scored based on the translated versions of their texts and the 4-5 selected quantitative and measurable writing parameters from the pilot. As an important and unique element in our scoring, we want to develop and indicate an uncertainty range that shows the teachers the unavoidable uncertainties in this type of complex assessment/scoring.

The model will be trained with simulated student text that has been created from a collection of students’ books and real data consisting of pairs of student texts and corresponding teacher texts that have been collected from our tool. We look to develop, implement, and scale our AI tool in the US market in the next 12 months.

Our AI assessment solution employs a sequence-to-sequence model leveraging the BART (Bidirectional and Auto-Regressive Transformers) architecture, a state-of-the-art transformer-based model known for its effectiveness in natural language processing tasks, including translation and text generation. This choice is particularly apt for our application, where we translate early-stage, phonetically spelled student writings into conventional text, facilitating the application of standard linguistic metrics for educational assessment.

Specifically, the AI component of our solution encompasses Natural Language Processing (NLP) and Machine Learning (ML), focusing on sequence transformation and text analysis. This AI framework is integral to translating unconventional early-stage writing into standard language forms, allowing for subsequent analysis using established linguistic metrics such as readability scores.

In terms of the technological stack, our system's front end is developed to integrate seamlessly with our current WriteReader platform, enhancing user interaction without disrupting familiar workflows. This front is built using modern web technologies, ensuring it is both responsive and accessible to users across various devices.

The model training happens in-house in collaboration with a professor and a PhD student from the Technical University of Denmark, allowing us full control over the model's training and deployment, ensuring that the model's learning and outputs are finely tuned to our specific educational context. This is crucial for maintaining the integrity and relevance of the feedback provided to teachers and students.

The model is deployed on an AWS Fargate setup as a FLASK app with privately accessible API-endpoints.

Additionally, we have integrated robust likelihood techniques to mitigate the effects of noise in our training data, enhancing the model's accuracy and reliability. This novel approach addresses the inherent challenges in analyzing early-stage writing, which is often noisy and highly variable.

Below is a sceeenshot of our model and attached is a published paper from our technical partner regarding the AI assessment module that we are in a process of testing with US schools that represent students from lower socio-economic background. 

Published article: WriteReader AI Assessment

Ensuring equity and combating bias in general and in this AI implementation, and particularly in educational settings, is central to our mission. Our approach incorporates several layers of strategic action to guarantee that our AI-powered writing assessment tool serves all students fairly.

1. Diverse Data Collection: We prioritize the collection of a diverse set of training data that represents the varied linguistic, cultural, and socio-economic backgrounds of students with access to our platform. This includes significant representation from public Transitional Kindergarten, Pre-K, and K-8 students from diverse demographics. By training our AI on a broad spectrum of early-stage writings, the model better understands and assesses the linguistic diversity present in student populations.

2. Inclusive Development Process: Our development team includes educators, software engineers, and AI experts from diverse backgrounds. This inclusion fosters a broader perspective during the tool's development and helps in understanding the unique challenges faced by marginalized student groups.

We actively engage with community stakeholders, including teachers and education experts from minority and low-income backgrounds amongst others, to gather insights and feedback on the tool's functionality and its impact on different student groups.

3. Equity-focused Features: The AI tool is designed to highlight educational disparities by providing teachers with data on where students need the most help, thus directing resources and attention to those who might otherwise be overlooked.

Features such as uncertainty range in scoring parameters also highlight the variability and potential subjectivity in assessing early-stage writing, making teachers aware of the need to look beyond mere scores when evaluating progress.

4. Training and Resources for Educators: We provide training for educators on how to use the AI tool effectively while being mindful of its limitations to ensure that they can fully leverage the AI tool to benefit their students.

5. Continuous Improvement and Feedback Loops: Our development cycle includes continuous feedback loops with users and stakeholders to refine the tool and its algorithms. This iterative process ensures that AI remains a dynamic and evolving solution capable of adapting to emerging educational needs and diversity challenges.

By integrating these practices, our AI tool not only adheres to MIT's definition of equity - providing access to opportunity and advancement for all - but also actively works to dismantle systemic barriers that have traditionally hindered the educational progress of marginalized groups.

4 Full-time employees 

4 part-time employees 

30 teacher ambassadors around the world 

We plan to start the implementation with Schools in the US. The schools represent students from lower socioeconomic backgrounds. There are 48 schools FARMS in the district of Montgomery that we work with. (Free and reduced-priced meals) Here is an outline of the implementation plan:

Plan and Timeline for Implementation of AI Assessment Module

Our teacher pricing is affordable for schools and we plan to give away the AI assessment module for free for the first year. Our goal is to find sponsors, foundations, etc. that can help us sponsor our development cost so that the schools can have free access. Our SaaS plan includes a free version of WriteReader. We will provide the AI module at a reduced cost to schools that don't have budgets. 

In our journey to address critical challenges, we encounter mostly barriers related to growth and finding relevant partners that can apply our technology.  Bridging communication gaps and understanding diverse perspectives is crucial for effective collaboration and our product development. Finally, in the market, reaching target audiences is a constant challenge.

We believe that partnering with Solve and the Bill & Melinda Gates Foundation can provide valuable support in overcoming these barriers. We would be looking for expanding our network through the help of the foundation. Financially, access to funding and strategic investments can fuel our expansion and R&D efforts.  In the market, leveraging Solve and the Foundation’s networks and resources can amplify our reach and impact, facilitating market penetration and growth. Through this partnership, we hope to address these barriers and advance our mission of creating positive change. We are planning to expand our presence in the US and hope that MIT and the Bill & Melinda Gates Foundation can help us increase our network and advice us on our GTM strategy.